Method and system of data modelling

• US 8,768,659 B2
• Filed: 09/18/2009
• Issued: 07/01/2014
• Est. Priority Date: 09/19/2008
• Status: Active Grant

• Current Assignee: University of Sydney
• Original Assignee: University of Sydney
• Inventors: Ramos, Fabio Tozeto, Nettleton, Eric, Durrant-Whyte, Hugh, Vasudevan, Shrihari
• Primary Examiner(s): Thangavelu, Kandasamy
• Application Number:
  US13/119,952
• Publication Number:
  US 20110257949A1
• Time in Patent Office: 1,747 Days
• Field of Search: 703/2, 703/6, 703/13, 435/6, 382/224, 382/154, 382/264, 382/132, 379/406.01, 706/16, 706/21, 706/25, 704/250, 702/19, 702/22
• US Class Current: 703/2
• CPC Class Codes: G06F 17/18 for evaluating statistical data , e.g. average values, frequency distributions, probability functions, regression analysis forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]

Abstract

A method for modelling a dataset includes a training phase, wherein the dataset is applied to a non-stationary Gaussian process kernel in order to optimize the values of a set of hyperparameters associated with the Gaussian process kernel, and an evaluation phase in which the dataset and Gaussian process kernel with optimized hyperparameters are used to generate model data. The evaluation phase includes a nearest neighbor selection step. The method may include generating a model at a selected resolution.

Claims

1. A method for modelling data based on a dataset, comprising:

in a computing system, executing;

a training phase, wherein the dataset is applied to a non-stationary Gaussian process kernel in order to optimize the values of a set of hyperparameters associated with the non-stationary Gaussian process kernel, and

an evaluation phase in which the dataset and the non-stationary Gaussian process kernel with optimized hyperparameters are used to generate model data, wherein the evaluation phase comprises a nearest neighbor selection operation, and wherein only a selected subset of nearest neighbor data in the dataset is used to generate each corresponding model data.

2. A method as claimed in claim 1, wherein the non-stationary Gaussian process kernel is a neural network kernel.

3. A method as claimed in claim 1, wherein the training phase comprises a combination optimization strategy.

4. The method of claim 3, wherein the combination optimization strategy comprises simulated annealing, with an output of a simulated annealing stage being input into a Quasi-Newton optimization stage.

5. A method as claimed in claim 1 wherein the dataset has at least 100,000 members, and wherein the method comprises sampling the dataset to obtain a training data subset that is applied in the training phase for optimization of the hyperparameters wherein the size of the training data subset is at least one order of magnitude smaller in size than the dataset.

6. The method of claim 1, wherein the dataset is measured terrain data.

7. The method of claim 1, comprising selecting an output resolution for the output model and generating the output model according to the selected resolution.

8. A method in accordance with claim 1, wherein the nearest neighbour selection operation utilizes a k-dimensional tree (KD-Tree) structure to access the dataset.

9. A method for modelling a dataset with spatial characteristics, comprising

deriving, using a processing unit, a model from the dataset by providing the dataset to a Gaussian process using a non-stationary kernel,

saving, within a memory, the dataset in a database that preserves the spatial characteristics of the dataset, wherein the database is adapted to provide a subset of nearest neighbour data in the dataset during regression of the Gaussian process.

10. A method in accordance with claim 9, wherein the database utilises a k-dimensional tree (KD-Tree) structure to store the dataset.

11. The method of claim 9, further comprising an operation of dividing the dataset into at least two subsets, the at least two subsets being used for deriving and applying the model.

12. A method as claimed in claim 9, wherein the non-stationary kernel is a neural network kernel.

13. A method as claimed in claim 9, further comprising:

in a computing system, applying a combination optimization strategy during regression of the Gaussian process.

14. A method of claim 13, wherein the combination optimization strategy comprises simulated annealing, with an output of a simulated annealing stage being input into a Quasi-Newton optimization stage.

15. A system for modelling an area of terrain, comprising

computer hardware, computer software, and computer memory containing information for defining a model for the area of terrain, the computer software comprising instructions for:

a modelling module arranged to receive measured terrain data and utilise a Gaussian process using a non-stationary kernel to derive a model for the area of terrain, and

a data structure arranged to receive a dataset comprising at least a subset of the measured terrain data, wherein the data structure preserves spatial characteristics of the dataset and is adapted to provide a subset of nearest neighbour data in the dataset during regression of the Gaussian process.

16. A system in accordance, with claim 15 wherein the data structure utilizes a k-dimensional tree (KD-Tree) structure to store the received data.

17. A system as claimed in claim 15, wherein the non-stationary kernel is a neural network kernel.

18. A system as claimed in claim 15, wherein the modelling module is further arranged to utilize a combination optimization strategy during regression of the Gaussian process.

19. A system of claim 18, wherein the combination optimization strategy comprises simulated annealing, with an output of a simulated annealing stage being input into a Quasi-Newton optimization stage.

20. A non-transitory computer readable medium storing executable instructions that when executed by a computer implement a method for modeling data based on a dataset, comprising:

a training phase, wherein the dataset is applied to a non-stationary Gaussian process kernel in order to optimize the values of a set of hyperparameters associated with the non-stationary Gaussian process kernel, wherein the non-stationary Gaussian process kernel is a neural network kernel, and

an evaluation phase in which the dataset and the non-stationary Gaussian process kernel with optimized hyperparameters are used to generate model data, wherein the evaluation phase comprises a nearest neighbour selection operation, wherein only a selected subset of nearest neighbour data in the dataset is used to generate each corresponding model data.